Repository navigation
feat(website): lift noindex on every canonical model page - #19405
MaanilVerma wants to merge 13 commits into
Conversation
…e top ten families Replace the all-or-nothing launch switch with a wave list keyed on Router model id, so the robots meta, sitemap and markdown twins follow the same set and survive the move to /hub/models. Wave 1 is the ten families with the most search demand; setting the launch to 'all' opens the rest.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Repository: Comfy-Org/ComfyUI_frontend/.coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
|
The latest updates on your projects. Learn more about Vercel for GitHub. 2 Skipped Deployments
|
QA for bcf5d6e, then a2621c0, 890eefb, 01ff3c8 and 1a0ac27 (the last row is the current state; earlier rows are superseded)One local build per side: before = #19399 head
Wave 1 per family (Router ids / pages): Krea 2 3/3, Seedance 2 4/11, FLUX 3 1/2, SeedVR2 and WaveSpeed upscalers 3/3, Ideogram 2/2, Wan 11/16, FLUX 2 and Kontext 4/4, Gemini Omni 2/6, Grok Imagine 4/8, GPT Image 5/5. Total 39 ids, 60 pages. Not covered here: the wave-1 pages on this stack still render the Grok showcase |
🌾 Fallow: ✅ PassedFallow audits the diff for duplication, complexity and dead code. It gates on new findings only, so when the audit completes, a red is something this PR introduced or moved — never inherited backlog. A red can also mean the audit itself did not finish; the details below say which happened. No new findings in the changed files. Reproduce locally# the base the gate compared against
npx fallow audit --root . --quiet --format json \
--changed-since e064209e621f14d46d803bf2f98d3ac37f75b2b8
Full docs: ADR-DEVEX-LINT-0015 🧩 Custom Nodes Ecosystem Matrix: 🟡 Withheld — could not measureExecutes the frontend JS of every registry pack that ships any, against this change, in both renderers. The eight This candidate changes nothing the matrix measures, so every expensive stage was skipped. Reproduce locally# build one shard's fixture, then run it
python3 scripts/registry-census/build_matrix.py --shard 1/4
MATRIX_RENDERER=vue MATRIX_VUE=1 \
npx vitest run --config vitest.matrix.config.mts
Full docs: scripts/registry-census/README.md 🌐 Website E2ETip All tests passed.
🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-19405.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-21d903v6y-comfyui.vercel.app Last updated: 2026-09-29T20:58:28Z for 📦 Bundle Size
|
Codecov Report✅ All modified and coverable lines are covered by tests. @@ Coverage Diff @@
## maanil/fe-2922-model-markdown-twins #19405 +/- ##
=====================================================================
Coverage 84.23% 84.24%
=====================================================================
Files 3210 3210
Lines 156400 157090 +690
Branches 40614 40734 +120
=====================================================================
+ Hits 131751 132340 +589
- Misses 22949 23049 +100
- Partials 1700 1701 +1
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 25 files with indirect coverage changes 🚀 New features to boost your workflow:
|
There was a problem hiding this comment.
🔍 Cursor Review — Consolidated panel
Triggered by @MaanilVerma.
Found 5 finding(s).
| Severity | Count |
|---|---|
| 🟡 Medium | 1 |
| 🟢 Low | 3 |
| ⚪ Nit | 1 |
Panel: 8/8 reviewers contributed findings.
… Router id set as a rollback lever
… with their own rollback switch
benjcooley
left a comment
There was a problem hiding this comment.
Comprehensive PR Review
Reviewed the diff against models-url-registry.ts, workshop-browse-content.ts, workshop-release.ts, routes.ts/hreflangAudit.ts, markdown-twins.ts, astro.config.ts and routes/models/[slug].astro. The switch itself is well shaped — the default-parameter seam is the right call, deciding by registry kind rather than URL shape is the right call, and keeping workflow pages behind their own boolean because 26 of 30 duplicate ranking /workflows pages is a correct read of the risk.
Verified as not problems, for the record:
- No hreflang fallout.
baseRoutes.workshop/modelsShowcaseare both/models, soisLocaleInvariantPathcovers every/models/**path andsupportsLocaleRoutestays false.isClusteredtherefore does not flip for the 148 new sitemap entries, socheck-hreflangwill not start demanding locale twins that were never built. - No path where
modelsUrlKind() === 'model'missesrouterIdByModelPage, today. Registrymodelsand the map are both derived fromworkshopModelswith the same/models/${slug}shape. - Registry
workflowentries are a superset ofworkflowModels(the registry uses every CLOUD/SERVERLESS display slug,workflowPagesForadditionally dropsunavailableand disabled ones), so the new workflow test is not vacuous and cannot pass by accident. - Reserved paths hold:
page.json,catalogue.json,showcase, apps and aliases all resolve to non-modelkinds and stay out.
Issue distribution
| Severity | Count |
|---|---|
| BLOCKER | 0 |
| SHOULD FIX | 3 |
| TRIVIAL | 2 |
The one that matters
The new tests call isIndexableModelPage with default arguments, which pins the shipped value of both switches. Setting launchedModelPages to a set — the emergency rollback this file documents — fails three assertions, and setting launchedWorkflowPages to true fails two more. So both documented operational levers require a test edit before CI will ship them. Given the repo rule against bypassing quality gates, that turns a one-line rollback into a code change made under incident pressure. Detail inline.
Second: routerId !== undefined on line 74 of indexing.ts turns a href/MODELS_BASE_PATH divergence into silent de-indexing of all 148 pages rather than a build error, and the "keyed on Router id rather than the URL" claim in the description only holds for the set, not for the lookup.
Third, non-code: the gate checklist should name the markdown twins. isExcludedFromSitemap feeds writeMarkdownTwins, so this flip also emits ~148 twins and llms-full.txt entries built from the showcase fallback HTML — same prerequisite as the <h1>, different artefact.
Nothing here blocks the design. Draft status plus the gate list is doing the right job; these are the seams I would tighten before the gate clears.
Automated review by Claude. If a comment is wrong, say why and I will drop it.
benjcooley
left a comment
There was a problem hiding this comment.
Requesting changes — upgrading my previous review from comment to a decision, per team convention. The design is right; two things need to change before the gate clears. Detail is in the inline comments on the earlier review.
1. The documented emergency rollback fails the test suite. The new tests call isIndexableModelPage with default arguments, so they assert the shipped value of both switches rather than the function's behaviour. Setting launchedModelPages to a set — the rollback model-page-launch.ts documents — fails indexing.test.ts lines 112, 113 and 118. Setting launchedWorkflowPages to true — the documented workflow launch — fails lines 112 and 143. Both operational levers now require a test edit before CI will ship them, which is the wrong ordering for an SEO kill switch pulled under pressure. Pass both switches explicitly in the behaviour assertions; the seam you added is exactly right, the tests just are not using it.
2. A href / MODELS_BASE_PATH divergence silently de-indexes all 148 pages. routerId !== undefined on indexing.ts:74 turns a mismatch between workshopModels[].href and the registry base path into false for every model page — noindex back on, all 148 out of the sitemap, twins no longer written, no error. Make the invariant loud at module load, or derive both from one expression.
Also worth folding into the gate checklist before merge (not a code change): this flip reaches writeMarkdownTwins through isExcludedFromSitemap, so it emits ~148 markdown twins and llms-full.txt entries from the showcase-fallback HTML. Same prerequisite as the <h1> fix, different and more durable artefact.
Verified as clear: no hreflang fallout (/models/** is locale-invariant, so isClustered does not flip for the new sitemap URLs), registry workflow entries are a superset of workflowModels so the workflow test is not vacuous, and every reserved path still resolves to a non-model kind.
Let isExcludedFromSitemap take both launch switches like isIndexableModelPage.
benjcooley
left a comment
There was a problem hiding this comment.
Approving. Both items I requested changes on are genuinely fixed, not papered over. Re-reviewed only 28309999e1 and 1a0ac27209 plus the files they touch.
1. Launch switches no longer pinned by the tests — resolved
1a0ac27209 passes both switches explicitly at every behaviour assertion, and it also gave isExcludedFromSitemap the same two optional parameters so the sitemap half of the lever is testable rather than just the noindex half. I walked every remaining default-argument call in the file (lines 35, 36, 37, 45, 47, 49, 76, 86, 91, 94) and each one lands on NOINDEX_PATHNAMES, MODEL_REDIRECT_PATHNAMES, isLegacyWorkshopRoute, or a non-Workshop path — so every one of them is independent of both switches by construction, not by luck. Flipping launchedModelPages to a set or launchedWorkflowPages to true can no longer fail the suite. The new sitemap assertion at line 138 closes the gap I flagged, and the alias check is now stronger than what I asked for: it passes 'all', true so aliases must stay non-indexable with both switches on.
2. Silent de-indexing — resolved, and the right way
routerIdsByModelPage checking modelsUrlPaths('model') against the href map is better than the size assertion I suggested, because it names the offending paths instead of just the count. Since both sides are the same cardinality by construction, the one-directional check is in fact a bijection check, so a base-path move in either file trips it. Confirmed the throw is environment-independent — no process.env anywhere in the workshop-browse-content / models-url-registry / availability chain — so it cannot fire only in one build. Also confirmed the new top-level modelsUrlPaths('model') call introduces no import cycle: workshop-browse-content reaches neither indexing nor models-url-registry, so modelsUrlRegistry is always initialised before indexing evaluates. And the routerId !== undefined guard that worried me is now required by the type system rather than load-bearing, since the invariant throws first.
The test at line 160 is the right test — it builds a /hub/models registry and asserts the throw, which is the actual FE-2972 scenario.
One consequence worth flagging to whoever does FE-2972: the base-path move must now change MODELS_BASE_PATH and workshopModels[].href in the same commit. Staging the registry first and the hrefs later will hard-fail the Astro config load. That is the intended trade, just worth knowing before someone splits that PR.
3. Markdown twins gate — resolved
The banner now names the 148 twins and llms-full.txt, and points at the same real-HTML PRs. Documentation was all I asked for.
Also good
Dropping the WORKSHOP_IN_BUILD stubs and collapsing the two Workshop-route tests into one is correct — isWorkshopRoute and isLegacyWorkshopRoute are both build-flag-independent, so the parametrisation was asserting nothing. vi is fully gone from the imports.
One open nit
The rollback test's derived expectation can go vacuous if that Router id leaves the catalogue — my suggested snippet caused that, and I left the detail and a one-line guard in the thread. Non-blocking.
Not verified
CI was still queued on 1a0ac2 when I looked, and this chat is read-only so I did not execute the suite. Everything above is from reading the code, not from a green run. Please confirm unit tests, typecheck and knip pass before merge — I did check that apps/website has no explicit knip project override, so the test-only import of routerIdsByModelPage counts as usage and should not trip the unused-export gate.
Approval covers the code. The draft banner and launch gates 2 to 8 still govern the merge.
…nto maanil/fe-2947-lift-model-noindex # Conflicts: # apps/website/src/config/indexing.test.ts # apps/website/src/config/model-page-launch.ts # apps/website/src/config/models-url-registry.ts
…ins' into maanil/fe-2947-lift-model-noindex
benjcooley
left a comment
There was a problem hiding this comment.
Approving again. The nit is fixed, and the more interesting event in this round is the base-branch merge, which I checked rather than waved through.
The nit — fixed
38762262a9 computes expected once and asserts not.toHaveLength(0) before the toEqual, so the rollback assertion can no longer pass vacuously. Exactly the guard I asked for.
The merge is the real change here, and it landed correctly
5d418474 pulled maanil/fe-2921-models-sitemap in, and base had meanwhile reshaped the two things this PR's new invariant is built on. Both adaptations are right:
ModelsUrlRegistrywent from a bareReadonlyMapto{ roots, entries }.modelsUrlPathscorrectly readsregistry.entries.values().buildModelsUrlRegistrygained a required secondrootsparameter. The drift test at line 173 correctly passes['/hub/models']— without itisUnderRootswould have thrown"outside"and the test would have passed for the wrong reason, asserting the wrong throw.- The invariant still holds, which was my main worry going in:
rootsis still[MODELS_BASE_PATH]andmodels:is stillworkshopModels.map(({ slug }) => slug), somodelsUrlPaths('model')still matchesworkshopModels[].hrefexactly. No build-time throw from the merge.
The integration risk I went looking for
Base's markdown-twin.test.ts mocked MODEL_PAGES_INDEXABLE — the export this PR deletes. Left alone that would not have failed cleanly: the mocked module would supply launchedModelPages: undefined, the default parameter would take undefined, and launched.has(routerId) would throw a TypeError rather than report a missing flag. The merge re-points the mock at both new switches with correct types. I checked the semantics too, not just that it compiles: listedInSitemap is derived live from the mocked value via default-argument isExcludedFromSitemap, so both halves of the assertion move together, and line 316 independently pins twin-presence to launched rather than just asserting self-consistency. That is the right shape for that test.
MODEL_PAGES_INDEXABLE is now gone repo-wide, and model-page-launch has exactly two consumers (indexing.ts and that mock), both adapted. No conflict markers anywhere in the touched files.
No scope creep
The /comfy-agent and /agent rows that appear in the range diff are base's, not this PR's — confirmed they are absent from the PR's own diff against d162b021. The PR's own footprint is still the five expected files.
Small bonus
Base's derived modelSlug / aliasSlug idiom replaced the hardcoded krea--krea-2-large entries in MODELS_PAGES_BY_KIND, so the catalogue-coupling I flagged is now gone at the table level as well as in the rollback test. The // Typed boolean so it can't narrow to false comment earns its keep — without it someone deletes the annotation and collapses the seam.
One forward-looking note, not a finding
Base introduced roots as an array. If FE-2972 ever registers model pages under two roots at once for the 308 transition, modelsUrlPaths('model') would return paths under both and this PR's invariant would throw, since href only covers one. assertOneHub blocks the naive two-root build today, so this is a heads-up for whoever does that move, not a problem in this diff.
Not verified
CI was still queued on 38762262 when I looked, and this chat is read-only, so I did not execute the suite. Everything above is from reading the code. Please confirm unit tests, typecheck, lint and knip are green before merge — the merge touched a shared test file, so a green markdown-twin.test.ts is the one I would specifically look for.
Approval covers the code. The draft banner and launch gates 2 to 8 still govern the merge.
|
Moved the base to #19404 (maanil/fe-2922-model-markdown-twins), so this PR's build now reflects the launch state: the empty-twin guard lives there. With the switch as shipped, the spinner-only model pages on this stack now get no twin and no markdown link instead of a "Loading" twin. Head 6e64cb7; proof is in the QA comment. |
benjcooley
left a comment
There was a problem hiding this comment.
Re-reviewed only what is new since 38762262: the restack onto maanil/fe-2922-model-markdown-twins (f4a9001f) and 6e64cb7c.
Restack is clean
Verified fe-2921 is an ancestor of fe-2922, and the new base tip d47506c5 is fully merged into the head — so this moved up the stack rather than sideways. Footprint is still the same five files, and the PR's own diff against the new base is unchanged except for markdown-twin.test.ts.
6e64cb7c is a faithful mock adaptation, and it caught the landmine a second time
The PR's own change to markdown-twin.test.ts is purely mechanical: MODEL_PAGES_INDEXABLE -> the two new switches, with true becoming 'all' and false becoming new Set() in both the beforeEach/afterEach pair and the parametrised case. Semantically equivalent in every spot.
Worth noting what it prevented. fe-2922 added a whole new describe('model pages') block that sets MODEL_PAGES_INDEXABLE = true — the export this PR deletes. Unadapted, that block would not have failed legibly: the mocked module supplies launchedModelPages: undefined, the default parameter accepts it, and launched.has(routerId) throws a bare TypeError. That is the second independent occurrence of the same integration hazard, and it was handled correctly both times. The mock stays complete (both runtime exports present, the type is erased).
One new finding
fe-2922's two model-page twin tests cover the spinner-only shape and the post-FE-2942 hero shape. Today's actual shape is neither, and it produces 148 near-identical Grok Imagine twins when this switch flips. Base also shipped the mechanism to prevent it. Detail and a one-line fix inline — non-blocking, and reasonable to land elsewhere in the stack.
Not verified
CI was still queued on 6e64cb7c; read-only chat, so I did not run the suite. All of the above is from reading code. The one I would specifically want green is markdown-twin.test.ts, since the restack rewrote a shared mock that two of base's new tests now depend on.
Approving. Draft banner and launch gates 2 to 8 still govern the merge.
…ins' into maanil/fe-2947-lift-model-noindex
benjcooley
left a comment
There was a problem hiding this comment.
Approving. My one open finding was wrong, and the new merge is clean. Nothing outstanding from me.
I was wrong about the markdown twins — withdrawn
I claimed the ModelsShowcase fallback would leak into 148 twins. It does not. I read slot="fallback" as an Astro island hydration fallback and never followed it into the Vue tree. It is a plain named slot: ModelsPage.vue:158-160 forwards it to WorkshopGate's #fallback, and WorkshopGate.vue:57 puts that behind v-else-if="view === 'denied'" chained off the view === 'loading' branch. During SSR useMounted() is false, so view computes to 'loading' at lines 30-31 and only #loading renders — WorkshopLoading, which carries data-twin-omit. The showcase markup is never emitted.
It also fails closed a second way: Astro serialises slots for hydrated islands into <template data-astro-template="fallback">, and TEMPLATE is in DROPPED_TAGS. Either path drops it.
The build numbers settle it empirically — 0 model .md files and 0 text/markdown links against 149 sitemap locs is exactly what an empty twin body plus the #19404 skip predicts. Thanks for measuring it rather than just asserting it; that was the right way to close the question. Thread resolved, no change needed.
The new merge is clean
afb640ba merges base again (d47506c5 -> cfd790ca). Verified it is a forward move rather than a force-push, and that the new base tip is fully contained in the head. The PR's own diff against the new base is byte-identical to what I approved last round: the same five files, the same 164/58.
The merge brought a large amount of unrelated base work (cinematic-studio, workshop-model-balance, workflow-render). None of it lands in this PR's footprint, and I checked the files this PR's invariants actually depend on — workshop-browse-content.ts, models-url-registry.ts, indexing.ts, markdown-twin.ts, integrations/markdown-twins.ts are all unchanged since 6e64cb7c, so the module-load Router-id invariant and the twin-skip path are exactly as verified.
The one file base touched that isExcludedFromSitemap reaches is workshop-release.ts, and the change is confined to assertWorkshopCloudEnvForBuild — an early-return removal and an error-message reword. isWorkshopRoute and isLegacyWorkshopRoute are untouched, so sitemap behaviour is unaffected.
Not verified
CI was still queued on afb640ba (17 queued, 2 success), and this chat is read-only, so I have not seen a green suite on any revision of this PR. Everything across my four passes is from reading code plus your build numbers. Worth confirming unit tests, typecheck, lint and knip before merge.
Draft banner and launch gates 2 to 8 still govern the merge. Approval is on the code.
…ins' into maanil/fe-2947-lift-model-noindex
benjcooley
left a comment
There was a problem hiding this comment.
Approving at b2611c22. One new commit, a base merge, and this one was not inert — it changed a mechanism this PR depends on, so I traced it.
What the merge brought
b2611c22 merges base again (cfd790ca -> 58805c39; verified forward move, not a force-push, fully contained in the head). Two base changes land in markdown-twins.ts:
TWIN_LINKloosened from/<link rel="alternate" type="text\/markdown"[^>]*>/to/<link\b[^>]*\btype="text\/markdown"[^>]*>/— now attribute-order agnostic.unlinkTwinnow throws instead of silently no-opping when there is no markdown link to remove.
That second one matters here, because this PR is exactly what routes 148 pages into that code path. Before the flip, model pages are excluded from the sitemap and continue at line 127, so unlinkTwin is never reached. After the flip they are indexable, so they fall through to the page.body === '' branch at line 150 and unlinkTwin runs on every one of them. If those pages did not emit a text/markdown link, the build would now fail 148 times.
Why it is safe
BaseLayout.astro:74 computes shouldNoindex = noindex || isNoindexPathname(pathname), and line 184 emits the twin link only when !shouldNoindex. For a model page isNoindexPathname is false and [slug].astro passes noindex={!isIndexableModelPage(pathname)}, so shouldNoindex reduces to !isIndexableModelPage. isExcludedFromSitemap reduces to the same thing, since isWorkshopRoute is true for kind model and the other three disjuncts are false. The two are the same predicate:
- indexable -> link emitted and not excluded ->
unlinkTwinfinds the link and strips it. No throw. - not indexable -> no link and excluded ->
continuebeforeunlinkTwin. No throw.
The dangerous combination (no link, but reaching unlinkTwin) is unreachable for model pages. The same holds when launchedWorkflowPages eventually flips, since both sides read the same function. And the assertion is a net positive for this PR specifically — it is precisely the desync that would otherwise ship 148 text/markdown links pointing at 404s.
PR footprint unchanged
The PR's own diff of the four config files is byte-identical to what I approved at afb640ba — 285 lines, no drift. markdown-twin.test.ts is still purely the mock adaptation; its diff only shifted because base inserted tests around it.
Worth noting base's two new tests ('drops the markdown link whatever its attribute order', 'fails the build when a contentless page has no markdown link to drop') both sit inside describe('model pages') and rely on this PR's beforeEach(() => { launch.launchedModelPages = 'all' }). That is now the third and fourth base test depending on the mock adaptation this PR made. It keeps working, but the stack is accumulating base tests coupled to this PR's switch — worth keeping in mind if the merge order ever changes.
Not verified
CI on b2611c22: 13 queued, 2 in progress, 3 success, 12 skipped, 0 failures. Nothing red, but still no unit-test, typecheck or knip result on any revision of this PR across five passes. Please confirm those before merge.
Draft banner and launch gates 2 to 8 still govern the merge.
…ins' into maanil/fe-2947-lift-model-noindex
benjcooley
left a comment
There was a problem hiding this comment.
Approving at 62c288f6. One new commit, again a pure base merge, and again it moved a mechanism this PR routes traffic into — so I re-checked the coupling I verified last round. Net effect: this PR got safer.
What changed
62c288f6 merges base (58805c39 -> e064209e; verified forward move, fully contained in the head). Two files, both base's:
unlinkTwin was reworked from "throw when there is no link to remove" to:
const unlinked = html.replace(TWIN_LINK, '')
if (unlinked.includes('type="text/markdown"')) throw ...
if (unlinked !== html) await writeFile(path, unlinked, 'utf8')That inverts the throw condition. A contentless page with no twin link is now a silent no-op instead of a build failure, and the throw fires only when a type="text/markdown" occurrence survives the replace — i.e. when TWIN_LINK under-matched.
Why this improves things here
Last round I verified the build was safe only because shouldNoindex and isExcludedFromSitemap both reduce to !isIndexableModelPage for model pages, so an indexable page always emits the link that unlinkTwin demanded. That coupling is no longer load-bearing for build success — if my analysis had been wrong, the old code failed the build 148 times; the new code just leaves the page alone. The coupling is now belt-and-braces rather than the thing standing between this PR and a red build.
The replacement guard is a real correctness check rather than a weaker one: it catches a twin link the regex failed to strip, which is exactly the dangling text/markdown pointing at a 404 that the loosened regex was introduced to prevent. For model pages the string appears exactly once, in the <head> link, lowercase, so it is matched, stripped, and the guard sees nothing left.
Two edges I looked at and am not flagging: TWIN_LINK has no g flag, so a page carrying two twin links throws rather than stripping both — correct, that is a bug worth failing on. And the guard's substring test could in principle fire on a page that merely mentions type="text/markdown" outside a link tag, but unlinkTwin only runs when <main> produced an empty body, so there is no prose or code sample to trip it.
PR footprint unchanged
The PR's own diff across all five files is hunk-for-hunk identical to what I approved at b2611c22 — 355 lines, the only delta being the blob hash on markdown-twin.test.ts because base edited around it. BaseLayout.astro (the link emitter), indexing.ts, workshop-browse-content.ts and models-url-registry.ts are untouched by this merge, so the module-load Router-id invariant is exactly as verified.
Base's two revised tests ('leaves a contentless page that never linked a twin untouched', 'fails the build when a contentless page keeps a markdown link it cannot drop') both sit inside describe('model pages') and rely on this PR's launchedModelPages = 'all' hook. That is the fifth and sixth base test coupled to this PR's switch.
Not verified
CI on 62c288f6 is fresh: 3 success, 12 queued, 0 failures, and the substantive jobs have not materialised yet. Across six passes I have still not seen a unit-test, typecheck, lint or knip result on any revision of this PR. On the previous head the 15 green checks were all infrastructure (Socket, Vercel, adr-links, binary-size, path filters) while build, lint, typecheck and repo-checks stayed queued indefinitely — worth confirming those actually dispatch rather than reading the green count as a pass.
Draft banner and launch gates 2 to 8 still govern the merge.
benjcooley
left a comment
There was a problem hiding this comment.
No code review this round — head is still 62c288f6, the commit I approved at 20:52, with no new commits, no force-push and no base change. Approval stands.
But I chased down why I have never seen a substantive check result across seven passes, and it is not slowness. Every base merge is cancelling the previous head's CI before it can finish, and the merges are arriving faster than the jobs take.
Run history for the four jobs that actually cover this PR:
20:48 62c288f6 pending CI: Website Unit
20:48 62c288f6 pending CI: Billing Web Unit
20:48 62c288f6 queued CI: Website Build
20:48 62c288f6 queued CI: Tests E2E
20:33 b2611c22 queued CI: Website Unit
20:33 b2611c22 queued CI: Billing Web Unit
20:33 b2611c22 cancelled CI: Website Build
20:33 b2611c22 cancelled CI: Tests E2E
20:20 afb640ba cancelled CI: Website Unit
20:20 afb640ba cancelled CI: Billing Web Unit
20:20 afb640ba cancelled CI: Website Build
20:20 afb640ba cancelled CI: Tests E2E
The base merges landed at 20:18, 20:32 and 20:46 — roughly one every 14 minutes, which is shorter than these jobs need. CI: Website Unit is the suite that covers all five files in this PR, and it has been cancelled or left queued on every single head. It has never produced a result. Seven of the GitHub Actions check suites on the current head still report runs=0.
The visible green count is misleading as a result. On b2611c22 the 15 successes were entirely infrastructure — Socket Security, Vercel preview, adr-links, binary-size, CLA, and the changes path filters — while build, lint, typecheck and repo-checks sat queued indefinitely.
Suggestion: hold off merging base for ~30 minutes and let one run complete on a stable head. Until CI: Website Unit goes green once, nothing in this PR has been verified by anything other than code reading — mine, plus your own local build numbers for the twins question. That matters more than usual here because base has reworked unlinkTwin twice in the last half hour and six base tests now depend on this PR's launchedModelPages = 'all' mock hook.
Not blocking, and not a code concern. Flagging it because "approved with a wall of green checks" reads as verified, and on this PR it is not.
## Summary **Deploy gate: comfy-router#46 is merged and live (2026-09-30). Its follow-up comfy-router#48 (serves `/hub/models.md` instead of 301ing it to a /workflows 404) must be deployed before this merges. After the merge, purge Cloudflare for `/hub/models*`, `/hub/workflows*` and the 30 workflow names in `hub-workflow-names.json`. The redirects ship as 307; FE-3083 flips them to 308 later.** This branch combines every Workshop /models SEO PR, so merging it to main replaces merging them one by one. It also launches: Comfy-Org#19405 commits `launchedModelPages = 'all'`, so all 148 model pages become indexable in the same deploy as the URL move. Part of FE-2912, FE-2942 (PR 2 is FE-3086), FE-2974 (/minimax 308 is FE-3087) Fixes FE-2978, FE-2989, FE-2991, FE-2990, FE-2959, FE-2981, FE-2977, FE-2979, FE-3014, FE-2936, FE-2980, FE-2975, FE-3085, FE-2972, FE-2935, FE-2950, FE-2940, FE-2930, FE-2985, FE-2957, FE-2961, FE-2986, FE-2921, FE-2922, FE-2947 ## Changes Router follow-up: comfy-router#48 (`/hub/models.md` routing, 324/324 router tests), deploy before this merges. The included PRs, in merge order. Every head was checked as contained at the final tip. | PR | Ticket | What | Ben | |---|---|---|---| | Comfy-Org#18887 | FE-2978 | Pricing Product markup and full video `uploadDate` | approved | | Comfy-Org#18889 | FE-2989 | A unique meta description on every model page | approved | | Comfy-Org#18890 | FE-2991 | Model page JSON-LD, emitted once noindex lifts | approved | | Comfy-Org#18892 | FE-2990 | Cleaner tag pills, guard against duplicate summaries | approved | | Comfy-Org#18893 | FE-2959, 2981, 2977, 2979 | Social tags, hub title, Organization data. The FE-2977 title was confirmed by the owner | approved | | Comfy-Org#19362 | FE-3014 | Every non-production build is noindex | approved | | Comfy-Org#19366 | FE-2936 | Model titles lead with "API & Playground" | approved | | Comfy-Org#19402 | FE-2980 | Every internal link ends with a slash, plus a CI check | approved | | Comfy-Org#19419 | FE-2975 | Footer links the catalogue on every page and locale | approved | | Comfy-Org#19476 | FE-3085 | `NOINDEX_ROUTES` decides noindex for serverless-animation | approved; re-approval pending on a one-line follow-up (CodeRabbit approved) | | Comfy-Org#19359 | FE-2974 | One typed redirect list that generates `vercel.json` | approved | | Comfy-Org#19369 | FE-2972 A | The old to new model URL map | approved | | Comfy-Org#18891 | FE-2935 | Model content shows whatever the flag says | approved | | Comfy-Org#18899 | FE-2942 | Model pages render in the static HTML | approved | | Comfy-Org#18898 | FE-2950 | Catalogue h1 and A to Z links in the static HTML | approved | | Comfy-Org#19375 | FE-2940 | Related models leave no page unlinked | approved | | Comfy-Org#19385 | FE-2930 | Examples show the prompt that made them | approved | | Comfy-Org#19394 | FE-2985 | Render guard e2e: no loader, no layout shift | approved | | Comfy-Org#19357 | FE-2957 | robots.txt stops blocking our scripts and styles | approved | | Comfy-Org#19360 | FE-2961 | Turning Workshop off keeps every Models page | approved (a CodeRabbit changes-requested review is still open) | | Comfy-Org#19386 | FE-2986 | One registry of every Models address | approved | | Comfy-Org#19421 | FE-2972 B1 | Model pages served at `/hub/models` | approved | | Comfy-Org#19422 | FE-2972 B2 | Every old model URL redirects (307) | approved | | Comfy-Org#19424 | FE-2972 B3 | Workflow pages at `/hub/workflows` with redirects | approved | | Comfy-Org#19399 | FE-2921 | One switch decides model robots and sitemap listing | approved | | Comfy-Org#19404 | FE-2922 | Clean markdown twins for model pages | approved | | Comfy-Org#19405 | FE-2947 | Lifts noindex on every canonical model page | approved | | comfy-router#46 | FE-3075 | Serves `/hub/models*` from the website and routes `/hub/workflows` by manifest. Merged and live | merged | These fixes exist only on this branch, because they only break once two PRs meet: - `scripts/check-models-html.ts` reads each page at `model.href` and skips models without a page (f1df1e4). - `e2e/fixtures/modelsCatalogue.ts` reads `dist/hub/models` (f87bb4c). - `routerIdsByModelPage` accepts an optional `href`, since Comfy-Org#19421 made it optional and Comfy-Org#19405 no longer typechecked (9783819). - `ModelStatus` links a successor only when it has a page, for the same reason on Comfy-Org#18899 (c4761d8). - The model breadcrumb in `modelJsonLd.ts` points at `/hub/models/` instead of the `/models/` redirect (eb985b6). Its label still says "Models" while the nav says "Hub". - Twin, meta-description and entry tests use the hub pages and the new tag set (86ac3b0, d4ffc0f). - The render guard finds the catalogue through `getRoutes().workshop`, and its `test.fail` is gone now that Comfy-Org#18898 renders the h1 (831499f). - A new test renders a real model page and checks its markdown twin has the model's h1 (9a6c5cb). - Main's newer catalogue and workflow e2e cases use the hub addresses (2aeda77). - `resetWorkshopAccountSource` resets `decision`; a clean auto-merge of main's Comfy-Org#19454 into Comfy-Org#18891 left it assigning a variable that no longer exists (c28c7a7). The conflict resolutions, briefly: `ALTERNATE_TWIN_SOURCES` stays deleted as Comfy-Org#18898 wants, so `/hub/models.md` comes from the real catalogue. The hub keeps Comfy-Org#18898's single static h1 instead of the `WorkshopHero` h1 that main's Comfy-Org#19320 added. `ExamplesTab` keeps Comfy-Org#19385's figure and caption with main's video poster. The Comfy-Org#19399/Comfy-Org#19405 indexing tests were rewritten for hub paths. The Comfy-Org#19359 redirect test no longer builds `/enterprise//` from slash-ended routes. The Comfy-Org#19369 URL-map test imports `PROVIDER_NAMES` from its new home. The full table of 47 resolutions is in the integration report. ## Review Focus - Launch rides along with the URL move. If launch should stay a separate step, set `launchedModelPages` in `apps/website/src/config/model-page-launch.ts` to `new Set()` before merging. - Main's Comfy-Org#19320 gives the catalogue island its own h1 through `WorkshopHero`. This branch keeps Comfy-Org#18898's static h1 and renders only the subtitle in `ModelsCatalogue.vue`, so the hub has exactly one h1. - Comfy-Org#19476 overlaps the open community PR Comfy-Org#18591 (same sitemap exclusion for serverless-animation). One of them should be closed once this lands. ## QA - [x] Gates: `pnpm typecheck` and website typecheck clean, `pnpm lint` 0 errors, eslint on `apps/website` clean, `pnpm knip` clean - [x] Website unit: 547 files, 10347 passed, 4 skipped - [x] Build A (`build:e2e`, CI env): 1193 pages; release gate, `check:hreflang`, `validate:llms-txt-links` (225 links), `validate:jsonld` (947 pages), `check:link-slashes`, `check:models-html` (148 pages) all pass - [x] Build B (`WORKSHOP_IN_BUILD=0`): same 182 Models pages as the on build, no `dist/workshop`, `/` is noindex; with the flag stubbed on, no Run, no Hub nav and no Firebase requests - [x] Build C (preview shape): 945 of 945 pages noindex, no canonicals except redirect stubs, no hreflang in the sitemap - [x] E2E on Build A: 644 of 645 passed; the one mobile failure passed on rerun. Earlier load flakes (navigation, demos, fdct, wan-3.0, network isolation, render guard) all passed when rerun alone - [x] Launch: 148 model pages indexable with canonicals, 149 hub entries in the sitemap, 148 non-empty twins with the model h1, 148 in `llms-full.txt`, 0 titles over 60, 30 workflow pages still noindex - [x] Redirects: 853 rules, 782 to /hub all 307, the MiniMax rows 307, every /hub destination built - [x] comfy-router#46 under `wrangler dev` in front of this build: 312 router tests pass; 148 old model URLs (both slash forms) 307 then 200; 30 workflow names served by the website and their old URLs 307; a non-manifest workflow 302 with no-store; `manifest.json` passes through; `/workflows/` and bare `/hub` unchanged - [x] After the router deploys and this merges: purge Cloudflare, then curl `https://comfy.org/hub/models/` for a 200 and three old `/models/<slug>/` URLs for one 307 each - [x] Rich Results Test on `/pricing/` and `/learning/animations/character-sheet/`, then "Validate fix" on Merchant listings and Videos - [x] Rich Results Test on two live model pages: breadcrumbs valid and the breadcrumb URL returns 200 with no redirect - [x] GPTBot and Googlebot curls on the hub and one model page, then a GSC live test on both - [x] Resubmit the sitemap in Search Console and URL-inspect about 20 model pages on day 0 and day 14 - Not covered: Rich Results and GSC need production; the redirect matrix used a local shim for `vercel.json`, since `astro preview` ignores it --------- Co-authored-by: GitHub Action <action@github.com> Co-authored-by: benjcooley <benjcooley@gmail.com> Co-authored-by: Deep Mehta <42841935+deepme987@users.noreply.github.com>
|
Superseded by the integration branch #19494, which combined this PR with every other Workshop SEO slice and has been merged to main. Closing this one. |
Warning
Draft on purpose. This is the launch, and it must not merge until items 2 to 8 of the launch gates are live and checked on prod: previews say noindex (FE-3014), robots.txt no longer blocks
/_website/(FE-2957), the registry and one redirect list (FE-2986, FE-2974),/hub/models/served by the website (FE-3075), final Hub URLs with one-hop 308s (FE-2972), and a Workshop-off build keeps every page (FE-2961). Model pages on this stack still render the Grok showcase<h1>, so the real-HTML PRs (#18891, #18899, #18898) and #19404 (model twins with the empty-twin guard) must be live first too. The flip also gives each model page a markdown twin in the section indexes andllms-full.txt. Until the real HTML lands, #19404's guard skips those twins rather than publishing empty ones, so the real-HTML PRs gate the twins as well as the<h1>.Summary
Every canonical model page (148) comes out of noindex at once and joins the sitemap and markdown twins. The 30 Models workflow pages stay noindex for now, behind their own switch, and apps, aliases, the showcase and reserved paths are never indexable.
Fixes FE-2947 (Linear)
Stacked on #19404
Changes
model-page-launch.tsswaps theMODEL_PAGES_INDEXABLEboolean forlaunchedModelPages, which ships as'all'.isIndexableModelPage()decides by registry kind, so it holds before and after the feat(website): move workflow pages to /hub/workflows (step B3): redirect every old URL (307 first, 308 after router #46, FE-3083) #19424 move to/hub/workflows/. Aworkflowpage followslaunchedWorkflowPages, which shipsfalse; amodelpage looks up its Router model id and checks it against the launch. The robots meta, the sitemap and the markdown twins already read this function, so all three move together.launchedModelPagesto a set of Router ids to keep indexed, or to an empty set for none. Then deploy. The set is keyed on Router id, but the lookup from page URL to Router id still reads the registry's model paths, so the FE-2972 move must keep them in step. If they drift, the build fails at module load instead of quietly de-indexing every page.launchedWorkflowPagestotrue, once they render real HTML andlegacyRedirectspoints the /workflows twins at them.Review Focus
isIndexableModelPage(pathname, launched, workflowsLaunched)andisExcludedFromSitemaptake both switches as default parameters, the same seammodelsUrlKind()uses for its registry, so the tests run over the real catalogue without mocking and stay green whichever way the switches are set.isIndexableModelPagekeeps its name although it now covers workflow pages too, because the FE-2922 twins branch on this stack calls it by that name. Happy to rename in a follow-up.